[Paper Review] Optimizing directed self-assembled morphology

Jian Qin, Gurdaman Khaira+5 more|arXiv (Cornell University)|Aug 2, 2013

Block Copolymer Self-Assembly1 references3 citations

TL;DR

This paper proposes a computational framework that combines the Cahn-Hilliard equation for simulating block copolymer equilibrium morphologies with covariance-matrix adaptation evolutionary strategy (CMA-ES) to optimize chemical substrate patterns for directed self-assembly. The method efficiently designs sparse, non-trivial patterns that guide lamellar-forming diblock copolymers into complex target morphologies—such as letters 'M' and 'E'—with minimal trial-and-error, achieving convergence in under 8 hours on a single processor.

ABSTRACT

Directed assembly of block polymers is rapidly becoming a viable strategy for lithographic patterning of nanoscopic features. One of the key attributes of directed assembly is that an underlying chemical or topographic substrate pattern used to direct assembly need not exhibit a direct correspondence with the sought after block polymer morphology, and past work has largely relied on trial-and-error approaches to design appropriate patterns. In this work, a computational evolutionary strategy is proposed to solve this optimization problem. By combining the Cahn-Hilliard equation, which is used to find the equilibrium morphology, and the covariance-matrix evolutionary strategy, which is used to optimize the combined outcome of particular substrate-copolymer combinations, we arrive at an efficient method for design of substrates leading to non-trivial, desirable outcomes.

Motivation & Objective

• To address the challenge of designing optimal sparse chemical patterns that guide block copolymers into non-trivial, aperiodic morphologies without direct geometric correspondence.
• To overcome the inefficiency of trial-and-error or random search methods in pattern design for directed self-assembly.
• To develop a systematic, computationally efficient approach for inverse design of substrate patterns that yield desired block copolymer morphologies.
• To demonstrate the method’s effectiveness on complex target patterns such as 'I', 'M', and 'E' using a minimal number of substrate spots.
• To enable scalable and parallelizable optimization for larger systems and diverse morphologies through a generic, extensible framework.

Proposed method

• Uses the Cahn-Hilliard (CH) equation to simulate the equilibrium morphology of diblock copolymers based on a Ginzburg-Landau free energy functional.
• Employs the Ohta-Kawasaki model to describe the free energy in the strong segregation regime, incorporating Flory-Huggins interaction parameter χ and degree of polymerization N.
• Applies the CMA-ES optimization algorithm to search for optimal substrate spot configurations that minimize the fitness function measuring deviation from the target morphology.
• Defines the fitness function as the L2-norm difference between the simulated equilibrium morphology and the target morphology in real space.
• Uses a fixed number of circular spots on the substrate, with positions as optimization variables, to achieve pattern interpolation.
• Enables parallelization of both the CH solver (due to matrix-vector operations) and CMA-ES (due to independent populations), enhancing computational efficiency.

Experimental results

Research questions

• RQ1 Can a computational optimization strategy efficiently design sparse chemical patterns that guide block copolymers into complex, non-periodic morphologies?
• RQ2 How does the CMA-ES algorithm compare to random or trial-and-error search in terms of convergence speed and solution quality for substrate pattern design?
• RQ3 To what extent can the Cahn-Hilliard model accurately predict equilibrium morphologies for complex target patterns under optimized substrate conditions?
• RQ4 How sensitive is the optimization outcome to initial configuration, and can ensemble averaging reduce this dependence?
• RQ5 Can the framework be extended to other morphologies and larger system sizes with scalable parallelization?

Key findings

• The CMA-ES algorithm successfully optimized substrate patterns to produce target morphologies such as 'I', 'M', and 'E' using only a minimal number of circular spots.
• Convergence to optimal solutions was achieved in approximately 8 hours on a single processor for modest system sizes (~2L₀ × 2L₀).
• The method demonstrated robustness across different target patterns, with similar convergence behaviors observed for 'I', 'M', and 'E' patterns under the same parameter set.
• The fitness function based on real-space morphology difference effectively guided the optimization toward low-error solutions.
• Multiple optimizations with different initial configurations yielded consistent results, suggesting low sensitivity to initialization when ensemble averaging is used.
• The framework is inherently parallelizable, with both the CH solver and CMA-ES population evaluations amenable to distributed computing for scaling to larger systems.